Due to factors such as open standard, high performance, low price, and being supported by many manufacturers, an Ethernet technology triumphs over technologies such as token-ring, FDDI (Fiber Distributed Data Interface), and ATM (Asynchronous Transfer Mode), and becomes a versatile interface for a LAN technology. With the emerging of new services and new applications using the Ethernet technology, how to ensure end-to-end QoS (Quality of Service) of a network application has become a biggest challenge faced by the Ethernet technology.
In order to solve this problem, the IEEE defines a congestion notification (CN) method in the 802.1Qau standard, so as to implement end-to-end traffic control of the Ethernet and improve an end-to-end QoS capability of the Ethernet.
In the CN method, first a congestion point (CP) and a reaction point (RP) in an Ethernet switch network need to be configured, as shown in FIG. 1, where a certain Ethernet switch in an Ethernet switch network is configured to a CP, and other nodes connected to the Ethernet switch are configured to RPs, and one CP corresponds to multiple RPs at the edge of the network. As shown in FIG. 1, a connecting line with two round endpoints between two nodes (for example, between an RP point and an RP point and between a CP point and an RP) in the figure represents a physical link between the nodes. In FIG. 1, a line with an arrow from a CP point to an RP point represents a transfer direction of control information.
For each data stream, a CP calculates feedback information (Fb) according to a queue length and a queue growth rate of an egress port of the data stream, and if Fb<0, it indicates that a queue congestion probability of the egress port of the data stream is increased, and the CP randomly selects a frame of packet from received subsequent packets of the data stream at a certain probability P, and sends control information to a RP that sends the fame of packet, where the control information includes information such as Fb and a congestion point ID; and after receiving the control information, the RP decreases a sending rate of the data stream according to a certain algorithm such as a method of decrement in proportion, and when no new control information is received in a set fixed period, increases the sending rate of the data stream, till an upper limit of the sending rate of the data stream is reached. The foregoing process is repeated continuously, thereby implementing network-level traffic control of the Ethernet.
The applicant finds that Ethernet traffic control performance of an existing Ethernet traffic control method is poor.